APT29’s OAuth persistence in Microsoft 365 hardens, pushing identity teams to clamp app consent and service principals
Marcus Webb ·
Security researchers report an active wave of intrusions that rely on malicious OAuth applications inside Microsoft 365 tenants, a tactic widely used by APT29, also called Midnight Blizzard, and now adopted by criminal groups. The method grants long lived access through Entra ID service principals and Graph API permissions like Mail.Read, Files.Read.All, Sites.Read.All, and offline_access, which survive password rotations and multifactor resets. Investigators say Wednesday’s docket of breach notices reflects this persistence method across multiple sectors.
The tactic matters because it shifts the control plane from endpoints to identity and authorization, which many organizations still treat as a one time configuration. Once an attacker lands delegated or application permissions, they can exfiltrate mailboxes and SharePoint or OneDrive data without interactive logins, and they can automate access using refresh tokens or app secrets.
Microsoft detailed Midnight Blizzard’s OAuth abuse in 2024 during the company’s corporate email breach review, noting admin consent grants and custom apps with high privilege Graph scopes. Proofpoint and others have since tracked similar abuse by Storm 0324 and Octo Tempest, where initial access came from social engineering or token theft, followed by the creation or consent of cloud apps to maintain access.
Defenders should set Entra ID application consent policies to block user consent by default, allow only verified publishers, and require admin approval for any app requesting high risk scopes. Enable the admin consent workflow, restrict multi tenant apps, and require publisher verification and app certification where feasible. Monitor for new service principals, new app role assignments, and credential additions, including certificates, client secrets, and workload identity federation.
Hunting should include Entra sign in and audit logs for non interactive service principal activity, Add service principal, and Add app role assignment to service principal events. In Microsoft 365, review Unified Audit Log for MailItemsAccessed and Graph access patterns at scale, and list Graph subscriptions that indicate automated data pulls. Detections should flag unverified or unknown publishers requesting Files.Read.All, Mail.ReadWrite, Sites.ReadWrite.All, and offline_access.
For remediation, remove the malicious enterprise application and service principal, revoke admin consent, delete app role assignments, and rotate or invalidate user refresh tokens. Use Revoke sign in sessions for affected accounts, invalidate app credentials, and purge hidden inbox rules and EWS or Graph subscriptions tied to the app. Validate Conditional Access coverage for service principals and enforce least privilege on app permissions.
Cloud identity abuse is now a primary initial access and persistence layer for both state aligned and criminal operators, which shifts defensive priority toward consent governance, service principal monitoring, and rapid app revocation workflows across all SaaS estates.